Image Acquisition:
For more information on sediment trap deployment and image acquisition, see the related dataset "Geltrap micrographs" (https://www.bco-dmo.org/dataset/749412). In brief, sediment trap collector tubes were deployed on various platform designs, including a neutrally-buoyant sediment trap (NBST), a surface tethered sediment trap (STST), and a Wire Walker (WW) trap. Trap tubes were filled with filtered water overlying a jar containing a polyacrylamide gel layer (Durkin et al. 2015). Upon recovery, collection tubes were allowed to settle for at least 1 hour before the overlying water was siphoned off. Jars containing polyacrylamide gel were removed from trap tubes and the remaining overlying water was carefully pipetted off the gel. Gels were stored at 4 degrees C and imaged within the following 2 days before being stored at -80 degrees C.

Polyacrylamide gel layers were imaged on a dissecting microscope (Olympus SZX16) with either a Luminera Infinity 2 (FK170124) or an Allied Vision Technologies StingRay (EN572 and EN581) camera attachment. Particles collected in gel layers during EN572 and EN581 were imaged under brightfield illumination. Particles collected in gel layers during FK170124 were imaged under both brightfield and oblique illumination, producing two separate sets of images for each sample. EN572 gel layers were imaged with a transparent grid to assist in tracking gel location during imaging. The grid was not used when imaging samples collected during subsequent cruises because the pronounced grid lines complicated image analysis. All gel layers were imaged at 4 increasing magnifications, though the combination of magnifications varied by cruise. To determine whether measured particle properties changed if gel layers are frozen, samples collected during FK170124 were thawed after being stored for approximately 1 year at -80 degrees C and imaged again under both brightfield and oblique illumination

Determination of Particle Flux:
Particles in gel images were quantified with an image processing protocol created using functions available in python’s Sci-Kit Image. Particles present in each micrograph were detected through a series of image transformation steps that maximize the detection of entire particle areas while minimizing the detection of imaging noise. The background was removed, a brightness threshold was applied, and an edge-detection kernel transformation was used to identify the in-focus particles. Duplicate particles detected in multiple focal planes were removed and the remaining particles were counted and measured.

Particles were categorized into 9 different sinking particle classes. Identities were assigned by manually identifying particle images. Aggregates were defined as loosely packed detritus with irregular edges. Dense detritus was defined as densely packed amorphous material, often brown or golden in color. Large, loose pellets were similar to dense detritus but were also elongated like a fecal pellet. Long fecal pellets were defined as long, thin, cylindrical fecal pellets with a smooth edge, such as the chitin-encased pellets produced by euphausiids. Short fecal pellets were also smooth-edged pellets but with an oval or ellipsoid shape, such as those produced by larvaceans. Mini pellets were defined as small (usually <100 micrometer ESD), spheres such as those produced by rhizarians and microzooplankton (Gowing & Silver, 1985). A small number of individual organisms considered to be passively sinking were also detected, including Rhizarians, primarily Phaeodaria, and various phytoplankton, usually diatoms. Pteropods, copepods, amphipods, foraminifera, and other zooplankton that probably swam into the gel and human-produced fibers were also detected but not considered passive flux and not included in the categories of sinking particles. Unidentifiable objects were also detected and were likely caused by out-of-focus particles, shadows, smudges, or noise detected by the image processing steps that are sensitive to the particular thresholds used.

Only identifiable, sinking particle classes from the dataset were considered in calculations of sinking flux. For each imaging magnification, particles were grouped by their equivalent spherical diameter (ESD) into logarithmically-scaled size bins. To calculate number fluxes at each magnification, the number of particles counted in each size bin was divided by the total imaged surface area of the gel and the trap collection time. Uncertainty of the number flux was estimated by applying these same calculations to the counting uncertainty (square root of the number of particles counted, ie. Poisson distribution).

The carbon flux by each particle type was determined by modeling particle volumes, calculating the carbon per unit volume, and multiplying the carbon per particle by the measured particle number fluxes in each size category.